HP 3245A Universal Source - 1.5ppm jump in output

[IanJ]

Well, this is wierd.....or is it?

I was running a 5hr test with my HP3245A, 3458A and R6581T.
Actually, I was testing the 10pF mod on the R6581T that MickleT came up with a day or so ago and posted.

Below is the playback of the 5hrs......notice the sudden 1.5ppm jump that the HP 3245A experienced.....and then drifted back down again!

The workshop aircon was not in it's tight metrology mode, and at 9pm switched off hence the temp stopped regulating and started to drop (winter here).

Coinciding with the 1.5ppm jump the workshop humidity appears to drop a little also, not sure if related.
For info, the PPM Dev. trace is on the R6581T.

My 3245A is modded, it's using an ADR1399 (3mA set) and with about 200hrs on it, and, I'm testing chopper op-amps on the Vref circuit.
The 3245A has also been off for the past few months and just switched back on about 50hrs ago.

I still have the original LM399AH so could go back to it for comparison.

Ideas?

Ian.

[TUMEMBER]

Too much convergence of the occurrence of the "spike". It could not have happened so simultaneously and identically on three meters. A software error, such an identical response to environmental conditions is unrealistic (different masses and delays for temperature/humidity change).

[IanJ]

Too much convergence of the occurrence of the "spike". It could not have happened so simultaneously and identically on three meters. A software error, such an identical response to environmental conditions is unrealistic (different masses and delays for temperature/humidity change).

If I read you correctly, the HP3245A is not a DMM......it's the 1Vdc source voltage feeding the two DMM's that I charted.

Ian.

[TUMEMBER]

HP3245A is a "voltage generator", but the DMM registers the change at its in
put. However, the 3458A and R6581T are not the same model of the meter and have different physical weights, and thus different delays for changes in environmental conditions. No time shift.

[iMo]

1399 is basically the same chip/technology as the 399. And many of them jump up/down from time to time..

[IanJ]

1399 is basically the same chip/technology as the 399. And many of them jump up/down from time to time..

Popcorn noise I've seen, but this latches for hours and slowly drifts back per the chart....which I wasn't aware of really. 1.5ppm seems a bit excessive to me.
I'll try the 399 back in and see how it performs in comparison.

Thanks,

Ian.

[iMo]

Do you have the 1u ser 5ohm snubber at the 1399 output??
399 does not need it, but 1399 does..

[chekhov]

I wonder if that could be some sort of mechanical issue, though if thinking about connections, it should not increase voltage.

Possibly instead of replacing one reference that maybe has popcorn with the one that has it with much higher probability it worth monitoring reference itself ?
Add wiring (to the ref or nearest buffer) and use one of the meters to monitor it, for example R6581 that has remove switching between front and back, or for example logging 2 channels of 3245A instead of only one and use 3458A for reference logging, so whatever options help to utilize your other meters more efficiently.

[IanJ]

I did have a snubber on the ADR1399, and I wonder if that's possibly where the issue lies. I had fitted a 0805 5ohm/1uF between pins 1 & 2 of the ADR, i.e. tented up between them to get them in series on the underside of the Pcb between pins 1 & 2.......so I wonder if that was causing a stability issue mechanically/thermally?

So, given that, I just swapped out the ADR for the original LM399 which I'd kept safe (removed the snubber and reset the zener current).

If it jumps with this then I'll go exploring with the 3458A in and around the reference circuit.
At the moment though I’m hoping to jump to a quick fix by trial and error because the jump issue can take hours/days to manifest.

Ian.

[aronake]

It seems this only happened once in one 5 h measurement or?

I would continue to measure for a couple of days (or longer preferably) to see if it was something recurring. If not, it may have been some kind of dramatic settling in event like some change in heat making something in the reference or other part of the circuitry "snap in place" after soldering.

[IanJ]

It is recurring, it’s not the first time I have seen it in my data.

Ian

[Kleinstein]

The jump is too large for the typical LM399 popcorn noise (usually some 0.5 ppm). From the curve it even looks more like 2.5 ppm than 1.5 ppm.
There are a few other parts that may show jumps. Some resistors, possibly as a defect can show jumps. Another point is mechanical stress, e.g. from slip stick movement at atachment points. The stress can than effect amplifiers and resistors.
Another point may be conncetors or switches / relays with no longer good contact.

Finding the source for such rather infrequent events is tricky. At least for the mechanical stress idea one could check the sensitivity by applying force while the unit is running.

[IanJ]

Been running the original LM399 for about 20hrs so far and it's been stable. Early days though.....

However, here's a wierd observation:
With the ADR1399 when I used to calibrate the 3245A (using WinGPIB/3458A) immediately after calibration and setting the 3245A to give a 1Vdc output (Hires mode) the 3458A would record 1.00000350 Vdc approx.
It would ALWAYS seem to output a slightly high reading despite having just calibrated it.....sometimes more, sometimes less than this figure.
But, calibrating with the LM399 fitted I got 0.99999978 Vdc.
So, I'm thinking there was maybe an oscillation on the ADR1399 (despite having the snubber) and when calibrating it would happen to catch at the lower voltage there which would render a higher voltage output after all the averaging etc in normal use.
Then in normal use the 'jump' I was experiencing could be the oscillation starting/stopping randomly.
Of course, I could be completely wrong......but it's a nice theory!

I'm logging data now from the 3458A/R6581T and hopefully if I can get 48hrs without the 'jump' I'll call it a fix.

Ian.

[coppercone2]

does it really need to be super low inductance?

I would get rid of the SMT parts and put TH parts there

[Kean]

I would get rid of the SMT parts and put TH parts there

Analog Devices use 0603 SMD parts for the snubber on their ADR1399 EVM.  A bad choice might be a cap with a low voltage rating or poor dielectric.  AD use 25V X5R from AVX.

[bastl_r]

Hi
If I switch on my (LED) lighting, I can forget about my long-term observations. What I'm trying to say is that there may have been an external interference.
Regards

[iMo]

Another diff except the snubber:
LM399 - 1mA
ADR1399 - at least 3.4mA zener current (will not work with 1mA)

[IanJ]

Hi all,

Here's 19hrs overnight with the LM399 fitted (snubber removed and 1mA zener current set).
It's a bit of a wild ride with the workshop temperature as it's not controlled overnight at the moment. Also, when it is controlled (per graph) it's not in tight metrology mode right now.

So far so good, no 'jump' in the 3245A experienced.

Ian.

[IanJ]

Another 14hr straight run, and no 'jump' recorded.
I'm gonna call this a fix for now as it would normally have glitched by now.
Putting it down to an ADR1399 problem (heater or diode), or the 'tight' 0805 snubber I'd fitted.
Either way, glad to have it up & running again.

I'll test the ADR1399 separately on the workbench when I get time.

Ian.

[aronake]

Another 14hr straight run, and no 'jump' recorded.
I'm gonna call this a fix for now as it would normally have glitched by now.
Putting it down to an ADR1399 problem (heater or diode), or the 'tight' 0805 snubber I'd fitted.
Either way, glad to have it up & running again.

I'll test the ADR1399 separately on the workbench when I get time.

Ian.

except for the jumps, did you see any, and if so approximately how much difference in short term noise using ADR1399 vs LM399?

[IanJ]

except for the jumps, did you see any, and if so approximately how much difference in short term noise using ADR1399 vs LM399?

I didn't exhaustively test, but I didn't see any difference between the two. The 3245A is a relatively noisy source, certainly in DCV mode so probably masked 'any' improvement.
Subject to testing (and ruling out the DAC), that's why on my unit the next aim is to replace the entire Vfer circuit and its amps/buffers if indeed I measure relative noise/drift on the Vref circuit.
If you look at the Pcb layout it is spread across a large area of the analog Pcb.

Ian.

[coppercone2]

I would get rid of the SMT parts and put TH parts there

Analog Devices use 0603 SMD parts for the snubber on their ADR1399 EVM.  A bad choice might be a cap with a low voltage rating or poor dielectric.  AD use 25V X5R from AVX.

he made a tent though, its not like placed on pads

[Kleinstein]

The noise of the 3245 does not look so bad. One can still see the 6581T to be a bit more noisy than the 3458. It also looks like some of the LM399 ref. jumps are visible.  There may be a slight improvement with changing some of the OP-amps (e.g. the TL072 at the fitler), but I would not expect that much of an improvement. The reference noise is likely a good part of the noise, so the step towards the ADR1399 could make some sense.

The snubber should not be that critical. A relatively small form factor MLCC could still use quite some of it's capacitance. So a 0603 may be a bit on the small size and may want some extra capacitance.  Ideally there would be some extra filtering between the reference and the amplifiers to reduce the noise from the 200 Hz (PWM frequency) band, that can enter the result even with zero output voltage.

A possible candidate to cause problems is also C405. Old electrolytic capacitors could show leakge and current spikes.  I would consider replacing it would a film capacitor or even MLCC with a bit higher capacitance.

[IanJ]

The 3245A is about twice as noisy as the PDVS2mini....so that's the angle I'm coming from.
A couple of screenshots attached for consideration........if I ever get around to it!

Room for improvement maybe?.....fun if nothing else.

Ian.

[Kleinstein]

From the last curve it looks like the noise has some odd component with a few points that read rather low. So not a more normal distribution, or the popcorn noise clearly visible.  Those rare low points could be related to switching transients and how many of them are in the integration time window.
There may also be some residual ripple from the PWM signal. A first step would be looking a bit more detailed at the noise. So is it more the low frequency or more the higher frequency noise or ripple that is a problem. Another point would be looking at the noise near zero and near 10 V to see more of the noise components.

The resistor values in RP100B between Pins 5,18,19 don't match with the 11.6 V ref. voltage given.

The weak points that I see are:
1) rather slow drivers (U110,U111) for the reference before the PWM switches. The switches cause quite some current spike and thus also voltage spikes there. Faster amplifiers with maybe lower open loop R_out could resuce these spikes. A first step could be some snubbers there (e.g. 100 ohm+10 nF).

2) a few rather noise amplifiers, TL072 (especially U117), LM358. These can contribute a little to the noise, though much of this would be higher frequency noise, above the filter frequency. Here the TL072 is actually not that bad. It is still a rather easy upgrade.

3) There is no filtering for the reference voltage. The PWM is between a positive and negative voltage and this mixing action brings some noise from the PWM frequency range to the near zero frequency. This part should be relatively easy to filter.

If not a low noise type the RP100 resistor network (specially the B part) could also contribute to the noise from current noise (e.g. an NOMCA network may be noticable)

[branadic]

Maybe I have missed it, but was R400 adjusted for operation with ADR1399?

-branadic-

[bastl_r]

What effect should the reduction have?
The +5V before CR400 is far too low anyway. 

[branadic]

Oh, I've read +15 V first.
Even with an LM399 that doesn't make sense and shouldn't work, unless that low potential is way lower than zero.

-branadic-

[Kleinstein]

The CR400 and R400 part is only for start up. During normal operation CR400 is blocking. No real need to adjust R400 as the exact value is non critical. The main point is avoiding that the OP-amp goes negative, which an OP07 with no negative supply can not do anyway.

[branadic]

Right, but R401 does need adjustment to provide 3 mA for ADR1399 (11.6 V - 6.95 V)/3.5k = 1.3 mA, which is not enough for it.

-branadic-

[IanJ]

Right, but R401 does need adjustment to provide 3 mA for ADR1399 (11.6 V - 6.95 V)/3.5k = 1.3 mA, which is not enough for it.

-branadic-

Yes, R401 was adjusted to set 3mA zener current when the 1399 was fitted.

Ian

[ch_scr]

I have seen 0805 "square block" smd resistors get "jumpy" after manhandling them a bit too much upon "desolder & put back". The MELF types seem more robust, bigger contact surface to the solder-able ends. Is the ADR1399 jumpy when you *now* measure it?

[Kleinstein]

Yes resistor can become "jumpy" and show extra 1/f type noise. This is not just a function of the form factor and soldering. Also normal THT resistors with careful sodering can show this. Not all resistors are equal in this respect and it is not so easy to predict which are good and which are a bit more noisy.
As an example I have found more noise in PTF56 type resistors (250 mW THT) than in some 0805 size SMDs (Susumu RR).

The HP3245A circuit for the 10 V range is however made in a way that only few resistors really contribute. The main point are 3 resistors inside the RP100 array. R400 is contributing too little to expect a noticible effect - unless the resistor is really defective or has a cold solder joint.
The rare jumps make it hard to say anthing definitive - it could be an odd reference.